This master’s thesis will investigate how a locally deployed AI assistant can be developed to analyze and answer questions about a large C++ codebase.
The underlying codebase comprises several million lines of C++ and is managed via Git. The goal is to enable developers to ask questions about the architecture, implementation, and dependencies of the existing code in natural language.
A key focus of the thesis is the investigation of suitable methods for code search and information retrieval for Large Language Models (LLMs), for example using
RAG, semantic search, and structural code analysis. The OpenWeight models used are to be run entirely locally on an NVIDIA DGX Spark, so that the source code does not have to leave the internal infrastructure.
In addition to the prototype implementation, various approaches regarding quality, accuracy, and practical usability will be evaluated and compared.
Your Profile
• Degree program with a focus on computer science or related fields, such as media informatics
• Strong analytical skills, an independent, results-oriented, and well-structured approach to work
• Proficiency in all Microsoft Office applications, as well as strong programming skills in C++
• Interest in modern AI methods
• Proficiency in German and English
We Offer You
• A collaborative work environment in a young team
• Exciting and challenging tasks with a high degree of autonomy
• The opportunity for a permanent position upon completion of your studies
• Flexible work hours tailored to your needs
